| Research Background:In the modern military environment,soldiers’ psychological states are directly linked to their task performance efficiency.Among these,rumination,a mental trait characterized by persistent focus on negative thoughts,significantly impacts psychological health and military performance.This trait has been proven to be closely related to various mental illnesses,including depression and anxiety disorders.Its excessive activity impairs individual psychological health and significantly affects soldiers’ daily functions and professional performance.Under the high-pressure military environment,the negative effects of rumination directly weaken the combat effectiveness of individuals and teams.Despite the critical importance of identifying rumination traits in psychological health assessments,existing methods mainly rely on subjective tools such as self-report questionnaires,which are susceptible to false answers,social desirability bias,and individual self-perception errors.This limits their effectiveness in precise screening and early warning.Rumination traits exhibit unique expressions in language,visual attention,and facial and physiological responses,stemming from its nature of persistent and repetitive negative thinking and indulgence.This thinking pattern not only leaves significant traces in verbal expression but also manifests in unconscious behavioral performance.Furthermore,individuals with high rumination display a different pattern of visual attention when exposed to stimuli,showing an excessive focus on negative information.The emotional fluctuations during the rumination process and their impact on the body—including changes in facial expressions and hemodynamics—provide specific physical and physiological evidence.Given this,the present study aims to explore more objective and scientific methods for identifying rumination traits.With the rapid development of Natural Language Processing(NLP)and artificial intelligence technologies,employing a multi-modal integration approach combining natural language with eye movement,facial action units(AUs),and facial blood flow to study rumination traits can capture individuals’ emotions,cognition,and attention from multiple dimensions.This overcomes the limitations of traditional methods and provides more precise and objective diagnostic information.Through the multi-modal data analysis of these technologies,not only can the accuracy of identifying rumination traits be improved,but a deeper understanding of the complex relationship between rumination and psychological health can also be achieved,providing a scientific basis for the screening,management,and intervention of military psychological health.Research Subjects and Methods:The study is divided into five parts,involving six specific experiments,with the subjects being recruits from the grassroots units.Part One: Based on natural language to compile rumination stimulus materials and verify their effects.Experiment one involved an initial screening of 4591 recruits using the Ruminative Response Scale(RRS),selecting 609 participants for one-on-one semi-structured interviews according to set inclusion and exclusion criteria.The interviews were recorded and transcribed into text,with 607 valid texts retained after proofreading and denoising.Experiment two involved 40 high-ruminators and 38low-ruminators rating the rumination stimulus materials across seven different dimensions on a 5-point scale.Through statistical tests of the ratings and comparisons of psychological states before and after the experiment,the effectiveness and reliability of the stimulus materials were assessed.Parts Two,Three,and Four: Based on single-modal data for the recognition of rumination natural language and model construction.A survey was conducted on 3373 recruits using the RRS,selecting 478 target individuals(271 high-ruminators,207low-ruminators)for the rumination stimulus experiment.Experiment three analyzed the visual attention patterns of participants through eye-tracking technology,constructing rumination identification models for the entire set of questions and individual questions;Experiment four utilized facial AUs data to analyze differences in facial expressions between the two groups under emotional provocation,constructing a rumination identification model;Experiment five explored differences in emotional reactions and physiological activation between the two groups through facial blood flow analysis,establishing a third set of single-modal recognition models.Finally,the complementarity of the three single-modal data types was analyzed to explore their mutual supplementation in rumination trait identification.Part Five: Based on multi-source synchronous multi-modal data fusion for rumination identification.Experiment six integrated the findings of single-modal data analysis from eye movement tracking,facial AUs,and facial blood flow,adopting a hierarchical fusion strategy combined with the XGBoost algorithm to construct a multi-data source rumination natural language recognition model.Research Results:Part One: Experiment one constructed a rumination corpus consisting of 607high-ruminator interview texts.A set of rumination natural language stimulus materials containing 37 passages and 17 rumination scenarios was developed based on this.Experiment two,in validating the stimulus materials through rating,found significant interaction effects between material type and personnel category via mixed-design ANOVA.Specifically,across all seven dimensions,the high-ruminator group rated the rumination materials significantly higher than the control group(repetitiveness: t=3.058,P<0.01;persistence: t=3.099,P<0.01;associativeness: t=3.189,P<0.01;vividness:t=3.336,P<0.01;uncontrollability: t=3.043,P<0.01;hypotheticality: t=3.413,P<0.01;representativeness: t=2.674,P<0.01),while there were no significant differences in neutral material ratings between the two groups.Parts Two,Three,and Four: Experiment three compared six types of eye movement indicators between the two groups when exposed to self-compiled rumination stimulus materials,showing significant differences in all indicators.A recognition model constructed with the Random Forest(RF)classifier achieved an accuracy of 73.22%,with an average accuracy of 69.60% for analysis by question;Experiment four analyzed facial AUs features of the two groups when exposed to stimulus materials,finding significant differences in activity,intensity,and variation.The final model’s classification accuracy was 62.75%,with an average accuracy increasing to 63.21% after analysis by question;Experiment five compared facial blood flow data between the two groups,finding significant differences in multiple facial areas such as the forehead,nose,and both sides of the face.An RF-constructed model achieved an accuracy of 60.64%,with an average accuracy of 58.03% for analysis by question.Additionally,the interaction between type of crowd and scenario type was significant,with blood flow differences under specific scenarios being particularly pronounced.Lastly,the complementarity analysis emphasized the high true positive rate(TPR)of 0.79 for eye movement data in identifying high-ruminators and the significant complementarity of facial blood flow data with facial AUs data in improving overall classification performance.Part Five: Experiment six,through hierarchical fusion analysis of multi-modal data using the XGBoost algorithm,achieved a classification accuracy of 87.03%,significantly higher than single-modal methods(eye movement 70.08%,facial AUs 64.85%,and facial blood flow 61.09%)and other comparison algorithms(RF 85.98%,KNN 85.35%,NB71.54%,and SVM 56.07%),proving the effectiveness of multi-modal fusion in improving identification accuracy.Research Conclusions:(1)This study successfully verified that stimulus materials can effectively initiate and measure the rumination levels of the target group.The results show that the self-compiled stimulus materials are effective in triggering and assessing rumination traits,confirming the potential application of this method in identifying psychological traits.(2)Through in-depth studies using eye-tracking,facial Action Units(AUs)analysis,and facial hemodynamics,this research revealed the applicative value of these non-invasive techniques in distinguishing between individuals with high and low levels of rumination.These single modality data showed certain effectiveness in differentiating individuals with varying rumination levels.Through a complementary analysis of these data,it was confirmed that the use of multimodal data fusion can significantly improve identification accuracy,compensating for the shortcomings of single modality approaches,and supporting the importance of multimodal fusion strategies in enhancing the precision of mental health assessments.(3)Multimodal data fusion significantly improved the accuracy of identifying rumination traits.By conducting a comprehensive analysis of various single modality data,this study confirmed the effectiveness of multimodal data fusion in enhancing diagnostic precision and emphasized the importance of integrating different types of data in psychological health assessments.(4)Specific contexts significantly influence individuals’ psychological and physiological responses,offering a new perspective on the complexity of rumination traits.This study,by analyzing the efficacy of stimulus materials in different scenarios,further reveals how environmental factors affect the expression of psychological traits.(5)By employing hierarchical fusion methods and the XGBoost algorithm,this study has significantly enhanced the efficiency and accuracy of identifying rumination traits.This approach effectively integrates information from multiple data types,optimizes the identification process,and achieves a high accuracy rate in psychological health assessments. |
PDF Full Text Request